Ways to Combine Fundamental and Market Data in Trading Models

Summary

The document outlines several broad ways to use slow-moving fundamental information alongside more frequent market data in a quantitative trading model. One option is to combine both sources in a single model and carry each fundamental observation forward until a newer one becomes available. The response cautions that a less predictive data source could affect how the combined model learns.

Alternatives include fitting separate models and combining their outputs, or using fundamentals to screen securities before making trading decisions with market data. These designs trade off simplicity and integration against possible loss of interaction effects or influence from weak predictors. The answer provides no empirical comparison, preprocessing details, or validation results, so it does not establish a best method. It emphasizes that suitability depends on the organization and strategy, and that approaches should be evaluated in their particular setting.

Key ideas

• A single model can combine data with different frequencies by carrying the latest fundamental observation forward.
• A weakly predictive data source may affect a model trained on combined inputs.
• Separate models can be combined through weighted outputs or another aggregation scheme.
• Fundamentals can screen securities before market data drives trading decisions.
• The answer offers alternatives rather than tested evidence for one universally best approach.

Tags

# Combine fundamental and market data into one ML model


# Combine fundamental and market data into one ML model












What are the best tested ways to preprocess data with very different frequencies such as fundamental and market data into same ML model for quant trading?

## Answer by Magic is in the chain (score 1)

https://quant.stackexchange.com/a/45402

There are many ways to go about this, for example:

You can build one global model using combined dataset (less frequent data will essentially carry the precious observed value until the next observation). Problem that you would face with this approach is the one of the data sources may be less predictive so you might not get a balanced model.

Build two separate models, and then combine them, either as a weighted average, or as a grid. Subjectivity and/or incorporation of less predictive attributes would be the major drawbacks.

Use the fundamental data to segment or short list the securities and then make trading based on the more frequent market data. Drawback would be the loss of interaction effects for example.

Can think of many others. The best approach would be pretty much driven by the organisation setup and the strategies so an approach that might work for one is not necessarily going to work for everyone, some one got to lose!

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.